41% of LinkedIn’s long posts are written by AI, and that’s the best news your writing has had in years

41% of the long posts on LinkedIn are fully written by AI. Not assisted, not touched up after. Written. That number comes from a study published on July 9 by Pangram, an AI-detection company, and it deserves more than the shocked repost it’s getting. Read closely, the study says something specific about where machine writing concentrates, who ships it, and why the people still writing in their own voice are about to be very easy to spot.

i.How they got the number

Most estimates of AI content crawl the open web, which mixes in millions of pages nobody reads. Pangram measured something narrower and more useful: what people actually see. Their browser extension scans posts as its users scroll, and users could opt in to share those scan results anonymously. Two months of that produced a dataset of 1,002,627 posts across LinkedIn, X, Reddit, Medium, and Substack. Every post counted once, nothing under 50 words scanned, all of it run through their latest detection model.

So this isn’t slop rotting in a dead corner of the internet. It’s the feed. The posts in front of you this morning.

Two honest caveats travel with the data, and the company itself makes the first one: people who install a detection extension are not a random sample of the internet, so Pangram’s CEO has framed the figures as a floor, not a ceiling. The second is that no detector is perfect, whatever its published error rate. Take the exact digits loosely. The shape of the finding is the point, and the shape is stark.

ii.Where people use it most: under their own name

Here is the finding that should have led every writeup. Across the five platforms, the place people were most willing to let AI speak for them was the one attached to their real, professional identity. LinkedIn made up about a third of everything scanned and produced 62% of all the content flagged as AI. Anonymous forums stayed mostly human. The professional network didn’t.

Think about what that means. On the platform where your name, your title, and your employer sit next to every word, most people decided the words were the part that didn’t need to be theirs.

Some of that is the platform’s own doing. LinkedIn built a “write with AI” button directly into the post box (since renamed, still there). Zero friction in, slop out. The company has noticed the reputation it’s earning; per the study, an executive announced that LinkedIn would start detecting and downranking generic AI posts, in an announcement that was itself flagged as AI-generated. The correction and the problem, sharing a byline.

That downranking matters more than the irony does. If the platform now suppresses generic machine writing, sounding like yourself stops being an aesthetic preference. It’s distribution.

iii.Longform is where it concentrates

Across all platforms, one in four posts over 250 words came back fully machine-written. On four of the five platforms, the longer the post, the more likely it was AI. The exception was Substack, where readers pay for writing, and where longer posts were actually slightly less likely to be machine-made. Audiences that pay for a voice appear to get one.

The comment sections tell the same story from the other side. On Reddit, replies came back 98.1% human, and that’s most of what Reddit is. The conversational corners of the internet are still people. But on LinkedIn, once you control for length, even the comments were slightly more likely to be AI than the posts. The slop isn’t only in the essays. It’s in the “great insights, thanks for sharing” underneath them.

Put the two findings together and the picture is precise: machine writing pools exactly where professionals perform expertise at length, under their own names. The long thoughtful post was the internet’s last costume for credibility, and it’s the first thing the machines learned to wear.

iv.Why this is good news for you

Here’s the part the shocked reposts skip. If almost half the long posts around yours are machine-written, the bar for standing out has never been lower. Not because you need to write better than you did five years ago. Because you need to write like a person at all.

A specific word an AI wouldn’t reach for. A sentence that turns where the reader didn’t expect. The small mundane detail only someone who was actually there would include. That texture used to be table stakes. In a 41% feed, it’s the whole game, and the readers scrolling past their fortieth identical post can feel it in under three seconds.

The wrong response is to reach for a detector, run your own drafts through it, and edit until the score looks safe. A detector can tell you a draft doesn’t sound like you. It has no idea what you sound like. That part was never the machine’s to fix.

The right response is cheaper and older: put your voice in before the draft exists, instead of scrubbing the machine out after. The tool that wrote 41% of the feed will write in your voice just as fast. It only writes in nobody’s voice because nobody told it otherwise.